Drug Resistance Analysis of Pancreatic Cancer Based on Universally Differentially Expressed Genes

Abstract

The high heterogeneity between patients can complicate the diagnosis and treatment of pancreatic ductal adenocarcinoma (PDAC). Here, we explored the association of universally differentially expressed genes (UDEGs) with resistance to chemotherapy and immunotherapy in the context of pancreatic cancer. In this work, sixteen up-regulated and three down-regulated genes that were dysregulated in more than 85% of 102 paired and 5% of 521 unpaired PDAC samples were identified and defined as UDEGs. A single-cell level analysis further validated the high expression levels of the up-UDEGs and the low levels of the down-UDEGs in cancer-related ductal cells, which could represent the malignant changes seen in pancreatic cancer. Based on a drug sensitivity analysis, we found that ANLN, GPRC5A and SERPINB5 are closely related to the resistance mechanism of PDAC, and their high expression predicted worse survival for PDAC patients. This suggests that targeting these genes could be a potential way to reduce drug resistance and improve survival. Based on the immune infiltration analysis, the abnormal expression of the UDEGs was found to be related to the formation of an immunosuppressive tumor microenvironment. In conclusion, these UDEGs are common features of PDAC and could be involved in the resistance of pancreatic cancer and might serve as novel drug targets to guide research into drug repurposing.

1. Introduction

Pancreatic ductal adenocarcinoma (PDAC) is considered one of the most aggressive malignant neoplasms, with a 12.8% five-year survival rate according to National Cancer Institute statistics. Complete surgery resection is the only possible cure for resectable PDAC, but most patients are already in an advanced stage at diagnosis [1]. Chemotherapy with/without radiotherapy has been a mainstay strategy for treating PDAC, but it only provides a modest improvement in survival [2]. Only 3% to 11% of patients have pathological complete responses after receiving chemotherapy with/without radiation therapy [3,4,5,6,7], which may be due to the high heterogeneity between patients and the ubiquitous drug resistance of PDAC. Immune checkpoint blockade (ICB) has changed the standard of care for multiple cancers and has brought hope to pancreatic cancer patients; however, almost all tentative trials have had little effect or failed before clinical application [8,9]. The most common reason for failure was the immunosuppressive tumor microenvironment (TME) resulting in low effector T cell infiltration, which is unfavorable for the immune response [10]. We suspect that there are some common features underlying the high heterogeneity of pancreatic cancer, which could be closely related to the mechanism underlying the ineffectiveness of current therapies. It is essential to identify and target these mechanisms to improve the early diagnosis rate and therapeutic efficiency.

Conventional differentially expressed gene (DEG) screening methods for bulk tissues only compare the average expression levels between disease and control groups, which cannot show the dysregulation status of each gene in a single disease sample or the frequency of differential expression of genes in all disease samples. The gold standard for individualized DEG screening is comparing the expression levels between paired tumor and adjacent normal samples from the same patient. However, the late diagnosis and difficulty of sampling pathologically normal samples in the complicated structure of the pancreas has led to few paired samples. To solve this problem, a robust individual DEG screening method, RankComp [11], has been developed to detect the dysregulation of genes in a single tumor sample relative to the normal background derived from normal samples from different studies. Thus, we can identify individual-level DEGs for unpaired tumor samples based on the aggregation of normal samples using this method. Here, we made full use of paired and unpaired PDAC samples to identify common molecular characteristics of PDAC patients to construct a robust early diagnosis model and determine the mechanisms underlying the resistance to current therapies.

In this study, using transcription profiles of paired and unpaired PDAC bulk samples, we identified individual-level DEGs in the PDAC samples and defined the genes that were dysregulated in at least 85% of the PDAC samples as universally differentially expressed genes (UDEGs). Bulk assays represent a population average, which masks the heterogeneity that exists at the single-cell level [12,13]. Thus, we analyzed the expression of the UDEGs at the single-cell level to explore their roles in the formation of tumors. Finally, we investigated the potential association between the UDEGs and overall drug resistance based on GDSC resistance data, as well as their correlation with responsiveness to immunotherapy using reported immunotherapy biomarkers.

2. Results

2.1. Identification of UDEGs in PDAC

The workflow for identifying UDEGs is shown in Figure 1A. In the 102 paired cancer and normal samples from GSE60646, GSE22780, GSE28735 and GSE15471, 180 genes were up-regulated and 32 genes were down-regulated in more than 85% of the tumor samples compared to the paired adjacent normal samples (Figure 1B). In the unpaired samples, based on 44,157,295 gene pairs that were stable in at least 90% of the 212 normal samples, DEGs were detected for each tumor sample using RankComp (FDR < 0.05). A total of 207 up-regulated and 153 down-regulated genes with dysregulated frequencies greater than 85% in 521 PDAC samples were detected. Finally, sixteen up-regulated and three down-regulated genes overlapped in the two DEG lists and were defined as UDEGs (Figure 1B–D).

2.2. Functions and Roles of UDEGs in PDAC

The P53 signaling pathway (p = 2.76 × 10⁻³) and cancer gene set from the CancerMine database (Hypergeometric test, p = 1.45 × 10⁻³) were enriched with UDEGs. Among them, SFN, LAMC2, GPRC5A, ANLN, ESM1, SERPINB5 and SLC6A14 have been reported to be closely related to the carcinogenesis and metastasis of PDAC (Appendix

Table A1. Functions of UDEGs in cancers.

Gene	Cancer Type	Role
CEACAM5	colorectal cancer	Driver
ESM1	esophageal cancer	Oncogene
SFN	prostate cancer, kidney cancer, pancreatic cancer, lung adenocarcinoma, malignant glioma, osteosarcoma	Oncogene
SFN	breast cancer, tongue squamous cell carcinoma	Tumor_Suppressor
LAMC2	pancreatic cancer, penis squamous cell carcinoma, esophagus squamous cell carcinoma, ovarian cancer	Oncogene
PCDH7	colon cancer	Oncogene
SERPINB5	breast cancer, stomach cancer, breast carcinoma	Oncogene
SERPINB5	breast cancer, prostate cancer, squamous cell carcinoma, melanoma, breast carcinoma, colorectal cancer, urinary bladder cancer, transitional cell carcinoma, stomach cancer, laryngeal carcinoma, lung squamous cell carcinoma, invasive lobular carcinoma, lung adenocarcinoma	Tumor_Suppressor
ANLN	pancreatic cancer	Oncogene
CORO2A	breast cancer	Oncogene
GPRC5A	lung non-small cell carcinoma, lung cancer, head and neck squamous cell carcinoma, stomach cancer, breast cancer	Tumor_Suppressor
GPRC5A	pancreatic ductal adenocarcinoma, pancreatic cancer, prostate cancer	Oncogene
ACADM	breast cancer	Oncogene

). SFN (Stratifin) is over-expressed and hypomethylated in cancer tissues compared with normal ducts and plays a role as an oncogene in pancreatic cancer [14,15]. LAMC2 (laminin subunit gamma 2) promotes cell proliferation, invasion and migration in multiple cancers, including pancreatic cancer [16,17,18]. Knockout of GPRC5A (G protein-coupled receptor family C, group 5, member A) reduced the proliferation and migration ability of pancreatic cancer cell lines and suppressed the resistance of pancreatic cancer cell lines to the chemotherapy drugs gemcitabine, oxaliplatin and fluorouracil [19]. The up-regulation of anillin (ANLN) promotes the progression and metastasis of PDAC [20,21,22]. ESM1 encodes a secreted protein that is mainly expressed in endothelial cells; it regulates the proliferation, migration, invasion and drug resistance of tumor cells, thereby promoting tumor progression and metastasis and playing an important role in the pathogenesis of pancreatic cancer [23]. Several reports have shown that the over-expression of SERPINB5 (serine protease inhibitor B5) can promote PDAC metastasis [24], and the marked up-regulation of SLC6A14 (solute carrier, family 6, member 14) in pancreatic cancer leads to worse survival [25,26,27]. Overall, these UDEGs are important in the carcinogenesis, invasion and metastasis of PDAC.

To further analyze the biological functions of the UDEGs, we generated a PPI network for the 19 UDEGs and found 155 proteins that interact with them (Figure 2A). Several carcinogenesis- and metastasis-related pathways, such as the cell cycle, focal adhesion, ErbB signaling and PI3K-Akt signaling pathways, are enriched with these 174 genes (Figure 2B). According to the Reactome pathway analysis, immune system, signal transduction, cell cycle, gene expression, disease, programmed cell death and extracellular matrix pathways were hierarchically enriched with the UDEGs (Figure 2C). These results indicate that in addition to cell proliferation and adhesion, changes in the immune system are also important for the carcinogenesis and development of pancreatic cancer.

2.3. Characteristics of UDEGs at Single-Cell Level

Because a bulk cancer tissue sample is a mixture of different types of cells [12,13], we explored whether the UDEGs are specifically dysregulated in cancer cells at the single-cell level. Here, we analyzed scRNA-seq profiles from CRA001160, which contains 24 PDAC and 11 control samples. From the tumor samples, 42,066 cells were acquired after quality control, which were separated into 39 clusters (Figure 3A), classified as acinar cells (cluster 20), B cells (clusters 7, 22 and 25), macrophages (clusters 1, 15, 32 and 37), endocrine cells (cluster 31), neuroendocrine cells (cluster 36), fibroblasts (clusters 5, 6, 23, 24, 33 and 38), endothelial cells (clusters 0 and 11), monocytes, stellate cells (clusters 2 and 4) and T cells. In particular, clusters 9, 10, 12, 13, 14, 16, 18, 19, 21, 26, 27, 28, 30, 34 and 35 not only expressed ductal cell markers (KRT19, MMP7, TSPAN8, SOX9 and LCN2), but also expressed cancer cell markers (MUC1 and PROM1). Thus, we defined these clusters as cancer-related ductal cells, and cluster 3, which only expressed ductal cell markers, was defined as ductal cells with normal characteristics.

From the 11 control samples, 15,553 cells passed the quality control and were classified into 27 clusters (Figure 3B), including acinar cells (clusters 7 and 15), ductal cells (clusters 0, 2, 4, 5, 6, 9, 11, 14, 16, 19 and 26), endocrine cells (cluster 20), endothelial cells (clusters 1, 8, 10, 17, 18 and 21), fibroblasts (cluster 3), stellate cells (cluster 12) and a small number of immune cells (T cells (clusters 22 and 24), monocytes (cluster 25), B cells (cluster 23) and macrophages (cluster 13)).

The major distinctions between the tumor and control samples were the large amount (27%) of cancer-related ductal cells and large proportion of immune cells in the tumor samples (25.29% in tumor samples compared with 4.15% in control samples) (Figure 3C,D). Although these cells are from merging different samples, the result still reflects the appearance of malignant lesions and an increase in inflammation in pancreatic cancer. In particular, up-UDEGs were specifically and highly expressed in cancer-related ductal cells while down-UDEGs had low expression (Figure 3E,F). Conversely, the down-UDEGs were expressed at high levels in the control samples, especially in some cell types, such as ductal cells, endothelial cells and fibroblasts. The KEGG enrichment analysis of specific highly expressed genes in each cell type found that most cells exhibit similar functions in both tumor and normal samples, such as T cells, macrophages or endocrine cells. However, multiple cancer-related pathways, such as necroptosis, pancreatic cancer, chemical carcinogenesis and reactive oxygen pathways, and metabolic pathways associated with proliferation, glutathione metabolism and glycolysis/gluconeogenesis (

Table 1. Functional enrichment analysis of different cell types in single-cell analysis of pancreatic cancer.

Cell Type	Number of Pathways			Consistance
	Normal	Tumor	Common
Acinar cells	5	6	3	55.00%
Cancer-related ductal cells	-	29	-	-
Ductal cells	35	30	22	68.10%
Endocrine cells	27	29	22	78.67%
Endothelial cells	94	82	64	73.07%
Fibroblasts	22	29	16	63.95%
Macrophages	70	74	64	88.96%
Stellate cells	39	28	23	70.56%
T cells	30	32	28	90.42%

) were enriched with the DEGs from cancer-related ductal cells, similar to the enrichment in bulk tissues. This result indicates that cancer-related ductal cells contribute most to the markedly high expression levels of up-UDEGs in the tumor tissues compared to the control samples and to the cancer-related functions in PDAC tissues.

2.4. UDEGs Involved in Resistance to Anticancer Drugs

Drug resistance is an important cause of poor chemotherapy responses in pancreatic cancer. We posited that the UDEGs we identified are involved in the universal drug resistance of PDAC patients. We downloaded the expression profiles of 32 untreated pancreatic cell lines and IC50 values of the 448 anticancer drugs used to treat the 32 pancreas cancer cell lines from the GDSC database. The 412 drugs that were used to treat more than half of these cell lines were retained. The mRNA expression profiles of cell lines were measured using the Affymetrix Human Genome U219 Array, which only contains nine of the nineteen UDEGs. Therefore, we only analyzed the relationship between these nine genes and drug resistance in our study. Based on the Spearman correlation analysis of the expression of the nine UDEGs and the IC50 for pancreatic cancer cell lines, 371 drug–gene relationships were found to be significantly correlated (Figure 4A, Appendix

Table A2. 371 significantly correlated drug–gene relations.

Drug	Gene	Threshold	Drug	Gene	Threshold
CAMPTOTHECIN	GPRC5A	Positive	OSU-03012	SERPINB5	Positive
VINBLASTINE	GPRC5A	Positive	EMBELIN	ANLN	Positive
VINBLASTINE	SERPINB5	Positive	EMBELIN	GPRC5A	Positive
CISPLATIN	ANLN	Positive	FH535	GPRC5A	Positive
CISPLATIN	GPRC5A	Positive	FH535	LRP8	Negative
CISPLATIN	SERPINB5	Positive	FH535	SDR16C5	Positive
CYTARABINE	ANLN	Positive	FH535	SERPINB5	Positive
CYTARABINE	GPRC5A	Positive	IPA-3	ANLN	Positive
CYTARABINE	SERPINB5	Positive	IPA-3	SERPINB5	Positive
GEFITINIB	GPRC5A	Positive	GSK650394	ANLN	Positive
NILOTINIB	GPRC5A	Positive	GSK650394	SERPINB5	Positive
OLAPARIB	ANLN	Positive	BAY-61-3606	SERPINB5	Positive
OLAPARIB	GPRC5A	Positive	THAPSIGARGIN	CORO2A	Positive
AZD7762	LRP8	Negative	BEXAROTENE	GPRC5A	Positive
AZD7762	SERPINB5	Negative	BEXAROTENE	SERPINB5	Positive
AZD7762	SFN	Negative	BEXAROTENE	SFN	Positive
AFATINIB	ACADM	Positive	BLEOMYCIN	ANLN	Positive
STAUROSPORINE	GPRC5A	Positive	BLEOMYCIN	SERPINB5	Positive
NUTLIN-3A (-)	GPRC5A	Positive	LFM-A13	GPRC5A	Positive
PD173074	ANLN	Positive	GW-2580	ANLN	Positive
PD173074	GPRC5A	Positive	PHENFORMIN	ANLN	Positive
PD173074	LRP8	Negative	PHENFORMIN	SERPINB5	Positive
PALBOCICLIB	GPRC5A	Positive	PAZOPANIB	CORO2A	Positive
PALBOCICLIB	SERPINB5	Positive	PAZOPANIB	LRP8	Negative
PD0325901	PCDH7	Positive	DACINOSTAT	ANLN	Positive
DASATINIB	PCDH7	Positive	EPOTHILONE B	ANLN	Positive
PACLITAXEL	ACADM	Positive	EPOTHILONE B	SERPINB5	Positive
PACLITAXEL	GPRC5A	Positive	TIPIFARNIB	ANLN	Positive
PACLITAXEL	PCDH7	Positive	AVAGACESTAT	ANLN	Positive
RAPAMYCIN	GPRC5A	Positive	AS601245	SERPINB5	Positive
RAPAMYCIN	LRP8	Negative	ISPINESIB MESYLATE	SERPINB5	Positive
RAPAMYCIN	PCDH7	Positive	TL-2-105	ANLN	Positive
SORAFENIB	ANLN	Positive	TL-2-105	SERPINB5	Positive
SORAFENIB	GPRC5A	Positive	TAK-715	ANLN	Positive
SORAFENIB	SERPINB5	Positive	BX-912	ANLN	Positive
IRINOTECAN	GPRC5A	Positive	BX-912	SERPINB5	Positive
GSK1904529A	GPRC5A	Positive	AS605240	SERPINB5	Positive
ERLOTINIB	CORO2A	Negative	GENENTECH CPD 10	SERPINB5	Positive
ERLOTINIB	SDR16C5	Negative	GSK1070916	ANLN	Positive
MK-1775	GPRC5A	Positive	GSK1070916	SERPINB5	Positive
MK-1775	LRP8	Negative	ENZASTAURIN	PCDH7	Negative
DINACICLIB	LRP8	Negative	GSK429286A	ANLN	Positive
GEMCITABINE	ANLN	Positive	GSK429286A	GPRC5A	Positive
GEMCITABINE	GPRC5A	Positive	GSK429286A	SERPINB5	Positive
TAMOXIFEN	LRP8	Negative	GSK429286A	SFN	Positive
EPZ004777	LRP8	Negative	QL-XII-47	SERPINB5	Positive
YK-4-279	GPRC5A	Positive	IC-87114	SERPINB5	Positive
YK-4-279	SERPINB5	Positive	UNC0638	SERPINB5	Positive
DAPORINAD	CORO2A	Positive	XMD14-99	ANLN	Positive
DAPORINAD	SERPINB5	Positive	XMD14-99	SERPINB5	Positive
TALAZOPARIB	ANLN	Positive	CP724714	ANLN	Positive
TALAZOPARIB	GPRC5A	Positive	CP724714	SERPINB5	Positive
XAV939	ACADM	Positive	JW-7-24-1	SERPINB5	Positive
DABRAFENIB	ANLN	Positive	NPK76-II-72-1	ANLN	Positive
DABRAFENIB	GPRC5A	Positive	NG-25	ANLN	Positive
DABRAFENIB	LRP8	Negative	TL-1-85	ANLN	Positive
TEMOZOLOMIDE	GPRC5A	Positive	ACY-1215	ANLN	Positive
TEMOZOLOMIDE	LRP8	Negative	ACY-1215	LRP8	Positive
IAP_5620	GPRC5A	Positive	ACY-1215	SERPINB5	Positive
AZD1208	GPRC5A	Positive	ACY-1215	SFN	Positive
EPIRUBICIN	ANLN	Positive	TUBASTATIN A	ANLN	Positive
EPIRUBICIN	GPRC5A	Positive	TUBASTATIN A	GPRC5A	Positive
CYCLOPHOSPHAMIDE	GPRC5A	Positive	TUBASTATIN A	SERPINB5	Positive
PEVONEDISTAT	ANLN	Positive	ZIBOTENTAN	ANLN	Positive
PEVONEDISTAT	SERPINB5	Positive	ZIBOTENTAN	GPRC5A	Positive
SAPITINIB	SDR16C5	Negative	ZIBOTENTAN	SERPINB5	Positive
SAPITINIB	SERPINB5	Negative	NSC-207895	ANLN	Positive
SAPITINIB	SFN	Negative	VNLG/124	SERPINB5	Positive
LCL161	SERPINB5	Negative	AR-42	ANLN	Positive
LAPATINIB	GPRC5A	Positive	AR-42	SERPINB5	Positive
LAPATINIB	LRP8	Negative	CUDC-101	ANLN	Positive
LUMINESPIB	CORO2A	Positive	CUDC-101	SERPINB5	Positive
ALPELISIB	ACADM	Positive	BELINOSTAT	ANLN	Positive
TASELISIB	ACADM	Positive	BELINOSTAT	SERPINB5	Positive
EPZ5676	GPRC5A	Positive	CAY10603	ANLN	Positive
IWP-2	LRP8	Negative	CAY10603	SERPINB5	Positive
LGK974	CORO2A	Negative	JNJ38877605	LRP8	Positive
WZ4003	LRP8	Negative	SU11274	SERPINB5	Positive
CZC24832	GPRC5A	Positive	KIN001-236	ANLN	Positive
CZC24832	LRP8	Negative	KIN001-260	ANLN	Positive
AZD5582	LRP8	Negative	NVP-BHG712	ANLN	Positive
GSK2606414	CORO2A	Negative	OSI-930	ANLN	Positive
PFI3	PCDH7	Positive	OSI-930	SERPINB5	Positive
PCI-34051	LRP8	Negative	CX-5461	ANLN	Positive
PCI-34051	PCDH7	Positive	CX-5461	SERPINB5	Positive
WNT-C59	SERPINB5	Negative	PHA-793887	SERPINB5	Positive
I-BET-762	ANLN	Positive	PI-103	CORO2A	Positive
RVX-208	PCDH7	Positive	FEDRATINIB	ANLN	Positive
GSK343	ACADM	Positive	Y-39983	ANLN	Positive
ML323	ANLN	Positive	Y-39983	GPRC5A	Positive
ML323	GPRC5A	Positive	Y-39983	SERPINB5	Positive
PRT062607	LRP8	Negative	YM201636	ANLN	Positive
AGI-6780	SFN	Negative	WYE-125132	PCDH7	Negative
CDK9_5576	LRP8	Negative	GSK690693	SERPINB5	Positive
CDK9_5038	LRP8	Negative	SNX-2112	ANLN	Positive
EG5_9814	GPRC5A	Positive	SNX-2112	GPRC5A	Positive
EG5_9814	LRP8	Negative	XMD13-2	ANLN	Positive
EG5_9814	SERPINB5	Positive	XMD13-2	SERPINB5	Positive
ERK_2440	LRP8	Negative	QL-X-138	SERPINB5	Positive
ERK_2440	SDR16C5	Negative	XMD15-27	ANLN	Positive
AZD5991	GPRC5A	Positive	T0901317	GPRC5A	Positive
AZD5991	LRP8	Negative	SELISISTAT	SERPINB5	Positive
PAK_5339	LRP8	Negative	KIN001-270	SERPINB5	Positive
TAF1_5496	ACADM	Negative	THZ-2-102-1	ANLN	Positive
TAF1_5496	ANLN	Positive	AT7867	SERPINB5	Positive
TAF1_5496	GPRC5A	Positive	AT7867	SFN	Positive
SELUMETINIB	GPRC5A	Negative	CI-1033	PCDH7	Negative
TENIPOSIDE	ANLN	Positive	TWS119	GPRC5A	Negative
TENIPOSIDE	GPRC5A	Positive	TWS119	LRP8	Positive
TENIPOSIDE	LRP8	Negative	TORIN 2	PCDH7	Negative
MITOXANTRONE	GPRC5A	Positive	PILARALISIB	SERPINB5	Positive
MITOXANTRONE	LRP8	Negative	GSK1059615	ANLN	Positive
DACTINOMYCIN	PCDH7	Positive	BRIVANIB, BMS-540215	SERPINB5	Positive
NELARABINE	ACADM	Positive	BIBF-1120	CORO2A	Negative
NELARABINE	GPRC5A	Positive	AST-1306	CORO2A	Negative
VINCRISTINE	GPRC5A	Positive	AST-1306	SDR16C5	Negative
VINCRISTINE	LRP8	Negative	APITOLISIB	LRP8	Positive
PODOPHYLLOTOXIN BROMIDE	GPRC5A	Positive	APITOLISIB	PCDH7	Negative
PODOPHYLLOTOXIN BROMIDE	LRP8	Negative	LIMK1 INHIBITOR BMS4	ACADM	Negative
DIHYDROROTENONE	ANLN	Positive	KB NB 142-70	CORO2A	Positive
ELEPHANTIN	LRP8	Negative	SPHINGOSINE KINASE 1 INHIBITOR II	SERPINB5	Positive
SABUTOCLAX	CORO2A	Positive	EEF2K INHIBITOR, A-484954	SERPINB5	Positive
MG-132	GPRC5A	Positive	ARA-G	GPRC5A	Positive
MG-132	PCDH7	Positive	C-75	SERPINB5	Positive
BDP-00009066	GPRC5A	Positive	CAP-232, TT-232, TLN-232	GPRC5A	Positive
BDP-00009066	LRP8	Negative	TRICHOSTATIN A	ANLN	Positive
BUPARLISIB	GPRC5A	Positive	TRICHOSTATIN A	GPRC5A	Positive
ULIXERTINIB	ANLN	Positive	PANOBINOSTAT	ANLN	Positive
ULIXERTINIB	LRP8	Negative	IMD-0354	SERPINB5	Positive
AFURESERTIB	GPRC5A	Positive	ETP-45835	SERPINB5	Positive
AFURESERTIB	LRP8	Negative	CD532	ANLN	Positive
AFURESERTIB	PCDH7	Positive	CD532	SERPINB5	Positive
AGI-5198	ANLN	Positive	ARRY-520	SERPINB5	Positive
AGI-5198	GPRC5A	Positive	SB505124	GPRC5A	Positive
AZD3759	SDR16C5	Negative	A-83-01	LRP8	Positive
AZD5363	GPRC5A	Positive	FTY-720	GPRC5A	Positive
AZD5363	LRP8	Negative	KOBE2602	SERPINB5	Positive
IPATASERTIB	LRP8	Negative	TRETINOIN	GPRC5A	Positive
IPATASERTIB	PCDH7	Positive	CI-1040	SDR16C5	Positive
GDC0810	GPRC5A	Positive	TEMSIROLIMUS	ACADM	Negative
GSK2578215A	GPRC5A	Positive	TEMSIROLIMUS	SDR16C5	Positive
GSK2578215A	LRP8	Negative	TEMSIROLIMUS	SERPINB5	Positive
GSK2578215A	PCDH7	Positive	BOSUTINIB	ACADM	Negative
I-BRD9	LRP8	Negative	AXITINIB	GPRC5A	Positive
I-BRD9	PCDH7	Positive	AXITINIB	SERPINB5	Positive
TELOMERASE INHIBITOR IX	PCDH7	Positive	GW441756	SERPINB5	Positive
MIRA-1	GPRC5A	Positive	VX-702	ANLN	Positive
MIRA-1	LRP8	Negative	ELESCLOMOL	ACADM	Negative
SAVOLITINIB	ANLN	Positive	ELESCLOMOL	SDR16C5	Positive
UMI-77	PCDH7	Positive	VISMODEGIB	ANLN	Positive
SEPANTRONIUM BROMIDE	ANLN	Positive	VISMODEGIB	SERPINB5	Positive
SEPANTRONIUM BROMIDE	CORO2A	Positive	BX795	SDR16C5	Positive
MIM1	GPRC5A	Positive	BX795	SERPINB5	Positive
BPD-00008900	GPRC5A	Positive	SL0101	ANLN	Positive
BPD-00008900	PCDH7	Positive	SL0101	SERPINB5	Positive
BPD-00008900	SERPINB5	Positive	JNK INHIBITOR VIII	LRP8	Negative
FORETINIB	ANLN	Positive	SB590885	ANLN	Positive
BIBR-1532	ANLN	Positive	CETUXIMAB	GPRC5A	Negative
PYRIDOSTATIN	GPRC5A	Positive	HG-5-88-01	LRP8	Positive
PYRIDOSTATIN	PCDH7	Positive	TW 37	SERPINB5	Positive
AMG-319	GPRC5A	Positive	TW 37	SFN	Positive
AMG-319	LRP8	Negative	XMD8-92	LRP8	Positive
MK-8776	GPRC5A	Positive	QL-VIII-58	PCDH7	Negative
VINORELBINE	ANLN	Positive	CCT-018159	CORO2A	Positive
VINORELBINE	GPRC5A	Positive	CCT-018159	SERPINB5	Positive
VINORELBINE	SERPINB5	Positive	CCT-018159	SFN	Positive
VX-11E	ANLN	Positive	QL-XII-61	GPRC5A	Negative
VX-11E	LRP8	Negative	IOX2	SERPINB5	Positive
LJI308	LRP8	Negative	(5Z)-7-OXOZEAENOL	SDR16C5	Positive
AT13148	GPRC5A	Positive	UNC1215	SERPINB5	Positive
AT13148	LRP8	Negative	UNC0642	CORO2A	Positive
GSK269962A	ANLN	Positive	UNC0642	SERPINB5	Positive
GSK269962A	GPRC5A	Positive	SGC0946	SERPINB5	Positive
GSK269962A	SERPINB5	Positive	ICL1100013	CORO2A	Positive
DOXORUBICIN	ANLN	Positive	ICL1100013	SERPINB5	Positive
DOXORUBICIN	SERPINB5	Positive	AZD4877	SERPINB5	Positive
ETOPOSIDE	ANLN	Positive	EPHB4_9721	ACADM	Positive
ETOPOSIDE	SERPINB5	Positive	EPHB4_9721	GPRC5A	Negative
MITOMYCIN-C	ANLN	Positive	EPHB4_9721	SFN	Negative
MITOMYCIN-C	SERPINB5	Positive	BPTES	LRP8	Positive
CP466722	SERPINB5	Positive	IGFR_3801	LRP8	Positive
PONATINIB	LRP8	Negative	JAK1_3715	SERPINB5	Positive
PF-562271	SERPINB5	Positive	PARP_0108	GPRC5A	Positive
DMOG	GPRC5A	Negative	PARP_9482	GPRC5A	Positive
OSU-03012	ANLN	Positive	TANK_1366	GPRC5A	Negative
OSU-03012	GPRC5A	Positive

).

Interestingly, most of the relationships between drug resistance and the expression levels of the up-UDEGs ANLN, GPRC5A and SERPINB5 were positive (Figure 4B), indicating that these three genes are closely related to the resistance mechanisms of pancreatic cancer. Moreover, we found that high expression levels of ANLN, GPRC5A and SERPINB5 were significantly associated with poor OS (overall survival) for TCGA PDAC patients (Figure 4C; ANLN: p = 0.034; GPRC5A: p = 0.045; SERPINB5: p = 0.028; log-rank test). Combining the above results, the universal drug resistance of patients with high expression of ANLN, GPRC5A and SERPINB5 might be one of the causes of poor survival. Thus, targeting these genes could be an effective way to suppress drug resistance and improve survival.

Among the significant correlations between LRP8 expression and the IC50s of anticancer compounds, 43/50 (86%) were negative (Figure 4B), indicating that most anticancer drugs exhibit higher efficacy when LRP8 is highly expressed; these drugs included tamoxifen (Rho = −0.49, p = 4.83 × 10⁻³), rapamycin (Rho = −0.51, p = 3.58 × 10⁻³) and afuresertib (Rho = −0.50, p = 5.39 × 10⁻³). Tamoxifen is widely used in the treatment of estrogen receptor-positive breast cancer and has been found to inhibit the myofibroblastic differentiation of pancreatic stellate cells (PSCs) and cell invasion into the tumor microenvironment [28], but it has not been used in the treatment of pancreatic cancer. Rapamycin is a lipophilic macrolide antibiotic that was originally developed as a fungicide and an immunosuppressant. Combined with cisplatin, it can affect the PI3K/AKT/mTOR signal transduction pathway, which leads to markedly increased cell apoptosis, indicating that rapamycin can mediate the sensitivity of pancreatic cancer cells to cisplatin [29]. Afuresertib is a novel inhibitor of the serine/threonine kinase AKT and has shown clinical efficacy as a monotherapy against hematological malignancies and could be used in combination with standard therapies for multiple myeloma [30]. In summary, our results identified some potential novel drugs for pancreatic cancer therapy that deserve further clinical research.

We also explored drug repurposing based on the UDEGs by generating a DGI network for the UDEGs and their PPI partners (Figure 4D). Some anticancer drugs (amatuximab, AZD-4877, ispinesib, letrozole, cergutuzumab amunaleukin, filanesib and nutlin-3), type II diabetes drugs (omega-3-carboxylic acids and ruboxistaurin) and drugs used to stimulate gastric and pancreatic secretions in hyperlipoproteinemia (levocarnitine) were found to interact with the UDEGs. Moreover, some novel drugs that have not been used to treat PDAC-related diseases were found to interact with the UDEGs, such as a drug for ocular hypertension (bimatoprost) and the antiepileptic drug losigamone. These results not only indicate that the UDEGs could be drug targets for pancreatic cancer but also provide new insights for drug repurposing to treat pancreatic cancer.

2.5. Roles of UDEGs in Response to Immunotherapy for PDAC Patients

The challenge in treating pancreatic cancer lies in its resistance to both chemotherapy and radiotherapy, as well as its unresponsiveness to the increasingly popular immunotherapy approach. Thus, we explored the roles of the UDEGs in the immune response to immunotherapy in PDAC patients.

A commonly used biomarker of immunotherapy, TMB, is universally low in PDAC patients, with an average of 0.73 mutations/megabase. For comparison, cancers with a high response rate to immunotherapy, such as melanoma and squamous cell lung cancer, have TMBs of 14.77 and 5.82 mutations/megabase, respectively (Appendix Figure A1A). Another marker, MSI-high, has been reported as a potential predictor of the effectiveness of immunotherapy in PDAC patients [31]. The TCGA mononucleotide and dinucleotide marker panel analysis found 24 PDAC patients with an indeterminate status, 115 with an MSS (microsatellite stable) status, 8 with an MSI-L (low MSI) status, and no patients with an MSI-H (high MSI) status (Appendix Figure A1B). Thus, it is difficult to apply TMB and MSI-H to measure immunotherapy responsiveness in PDAC.

The TME of PDAC is changeable, and modulation of the cell state can strongly influence drug responses, with state-specific vulnerabilities demonstrated in vivo and ex vivo [32]. Thus, we identified the proportions of 22 tumor-infiltrating lymphocytes in the TME of the TCGA PDAC patients using the CIBERSORT deconvolution algorithm. Consistent with previous studies, pro-tumor immune cells such as M0 and M2 macrophages and CD⁴⁺ memory resting cells were relatively abundant, but effector T cells such as CD⁴⁺ T cells and CD⁸⁺ T cells were relatively rare (Figure 5A), indicating a strong pro-tumor property and an immunosuppressive TME in PDAC. The results showed that most up-regulated UDEGs were significantly negatively correlated with the proportion of TME components associated with the response to ICB, including CD⁸⁺ T cells, naive B cells and activated dendritic cells, and significantly positively correlated with the proportion of TME components associated with resistance to ICB, including M0 macrophages, resting dendritic cells and Tregs (Figure 5B). Moreover, the down-UDEGs CYB5A and EPHX1 were significantly positively correlated with the proportion of CD⁸⁺ T cells, which have been reported to be associated with ICB responsiveness (Figure 5B). These results suggest that high expression of the up-regulated UDEGs and low expression of the down-regulated UDEGs are related to the formation of an immunosuppressive TME and might predict an unfavorable response to immunotherapy.

3. Discussion

The high heterogeneity between patients and the ubiquitous drug resistance are the main reasons for the poor efficacy of the current therapies for pancreatic cancer. Understanding the common molecular characteristics is essential to understanding the mechanisms of this universal therapeutic resistance and might be the key to overcoming this resistance. In this work, sixteen up-regulated and three down-regulated UDEGs were identified using paired cancer and normal samples and unpaired cancer bulk tissues. The single-cell analysis validated the high expression levels of the up-UDEGs and low levels of the down-UDEGs in cancer-related ductal cells compared to those in the control samples, which could represent the malignant changes in pancreatic cancer.

Among the nineteen UDEGs, other than the seven UDEGs reported to be carcinogenic and prometastatic genes involved in pancreatic cancer, the other UDEGs, such as ESM1, PCDH7, CORO2A, CEACAM5, GALNT5 and TMPRSS4, are also involved in other cancers. ESM1 (endothelial cell-specific molecule 1) can promote cancer progression and metastasis through the regulation of tumor cell proliferation, migration, invasion and drug resistance. PCDH7 (protocadherin 7) was reported to be related to cell growth, development and progression in prostate cancer [33], cervical cancer [34] and breast cancer [35]. CORO2A (coronin 2A) plays a critical role in cell migration and proliferation in breast cancer [36]. CEACAM5 (carcinoembryonic antigen-related cell adhesion molecule 5, also known as CEA) was reported to be a driver gene in colorectal cancer [37,38] and is related to tumor differentiation, invasion and metastasis [39]. GALNT5 (GalNAc transferase 5) mediates the carcinogenesis and progression of cholangiocarcinoma [40] and gastric cancer [41]. LRP8 is over-expressed in multiple cancers and is a potential therapeutic target [42,43,44,45]. TMPRSS4 is an important mediator of cell migration, invasion, epithelial–mesenchymal transition and metastasis in colon cancer cells, and increased TMPRSS4 expression correlated with colorectal cancer stage progression [46].

Based on these UDEGs, we performed a drug resistance and immune infiltration analysis and investigated the correlation between the expression of the UDEGs and the response to treatments for PDAC. These UDEGs were found to be closely related to the ubiquitous resistance of pancreatic cancer. For example, the higher expression of ANLN and GPRC5A was correlated with higher resistance to gemcitabine in pancreatic cell lines (ANLN: R = 0.43, p = 1.5 × 10⁻²; GPRCA5: R = 0.53, p = 2.3 × 10⁻³; Spearman correlation, Appendix Figure A2A). In contrast, in the TCGA PDAC cohorts, the non-responsive patients tended to express higher levels of ANLN and GPRC5A than the responsive patients (Appendix Figure A2B). Thus, PDAC patients with higher levels of ANLN and GPRC5A expression are more likely to be resistant to gemcitabine at the cell and tissue levels. Moreover, previous studies have shown that a down-regulation of ANLN can enhance the sensitivity of pancreatic cancer cells to gemcitabine [47] and inhibit doxorubicin resistance in human breast cancer cells [48]. Another study proved that knocking out GPRC5A suppressed the resistance to gemcitabine in a pancreatic cancer cell line (MIA PaCa-2) [19], and its over-expression enhanced the resistance of ovarian cancer cells [49]. SERPINB5 over-expression has been shown to promote metastasis in pancreatic ductal adenocarcinoma, and knockdown of SERPINB5 reduced the primary tumor weight and significantly decreased the amount of metastasis [24]. However, it has not been validated to be related to drug resistance and needs further validation through experiments. These results indicate that the UDEGs could be potential drug targets for pancreatic cancer. Thus, we also performed a DGI network analysis and found that some drugs used in other diseases interact with the UDEGs or their PPI partners that play important roles in the treatment of pancreatic cancer. These drugs are approved by the FDA and have passed preclinical testing, which not only ensures their safety but also greatly shortens the time for drug development and reduces the risk of failure.

In conclusion, we identified 19 universally differentially expressed genes (16 up-regulated and 3 down-regulated) in PDAC through a combination of bulk and single-cell analyses. These UDEGs are closely related to the drug resistance of PDAC, providing several potential targets for repurposing anticancer drugs to treat PDAC.

4. Materials and Methods

4.1. The Sources and Preprocessing of Expression Profiles

All the data used in our study were downloaded from public databases. Transcription profiles of PDAC and adjacent normal samples from PDAC patients and normal samples from pancreatic disease-free people came from the GEO (ncbi.nlm.nih.gov/geo, accessed on 13 March 2022), TCGA (cancergenome.nih.gov, accessed on 23 March 2022), GTEx (The Genotype-Tissue Expression, gtexportal.org, accessed on 13 March 2022) and ArrayExpress (ebi.ac.uk/arrayexpress, accessed on 13 March 2022) databases (

Table 2. Overview of samples used in this study.

GEO Accession	Platform	Sample Size
		PDAC	PDAC_Adjacent	Normal
Unpaired Samples
GSE62452	GPL6244	69	61	0
E-MTAB-1791	A-MEXP-2271	268	74	41
GSE91035	GPL22763	25	0	8
GSE62165	GPL13667	118	13	0
GSE56560	GPL5175	28	4	3
GSE71989	GPL570	13	0	8
Total		521	212
Paired Samples				-
GSE60646	GPL5175	10	10	-
GSE22780	GPL570	8	8	-
GSE28735	GPL6244	45	45	-
GSE15471	GPL570	39	39	-
Total		102	102	-

PDAC: pancreatic ductal adenocarcinoma.

).

For the data obtained using the Affymetrix platform, we applied the robust multi-array average algorithm for background adjustment without inner-sample normalization. Every probe was mapped to an Entrez gene ID according to the corresponding platform annotation file. The probes that mapped to multiple genes were discarded. If multiple probes were mapped to the same gene, the expression value of this gene was defined as the arithmetic mean of the values of the mapped probes. For the data sets obtained using the Illumina and Agilent platforms, we directly downloaded the processed expression data. For the profiles from the TCGA database generated using RNA-seq, the level 3 data were directly downloaded. Then, the Ensemble gene IDs were mapped to the Entrez gene IDs of protein-coding genes and used for further analysis. In this work, the PDAC and adjacent normal samples from the same patient were called paired samples; otherwise, they were referred to as unpaired samples.

4.2. Identification of Universally Dysregulated Genes in PDAC

We screened UDEGs using both paired and unpaired PDAC samples from public databases. In the paired samples, if the expression of a gene in the tumor sample was higher than that in the paired adjacent normal sample, it was defined as up-regulated and vice versa. For the unpaired samples, we screened dysregulated genes in individual samples by applying the RankComp method [11]. First, the normal background was derived from stable gene pairs, namely those gene pairs in which the rank of one gene is higher than or lower (G_i < G_j or G_i > G_j) than that of the other gene in more than 90% of the normal samples from different sources. Next, reverse gene pairs were defined for each disease sample as gene pairs with reversed rank ordering in comparison with their ordering in normal samples (G_i > G_j → G_i < G_j or G_i < G_j → G_i > G_j). Afterwards, Fisher’s exact test was used to test the null hypothesis that the numbers of reverse gene pairs supporting the up-regulation and down-regulation of a given gene (G_i) in a given disease sample (k) are equal. Based on the ratios of gene pairs with different ordering patterns in the normal and disease samples, we determined whether G_i is up-regulated (if the ratio in the disease sample is greater than in the normal samples) or down-regulated (if the ratio is smaller) or has stable expression (if the ratios are equal). A gene was determined to be a differentially expressed gene only if it remained significant after excluding the down-regulated (or up-regulated) partner genes from the reverse gene pairs supporting its up-regulation (or down-regulation) designation. Finally, the dysregulation frequency of each gene in the two approaches was calculated, and the overlapping genes with more than 85% frequency in both DEG lists were retained as UDEGs for the subsequent analyses.

4.3. Protein–Protein Interaction Network

A protein–protein interaction (PPI) network was created using PPIs from the STRING database [50]. Each PPI is annotated with a confidence score, which represents how likely STRING judges an interaction to be true given the available evidence. The scores range from 0 to 1, with 1 representing the highest possible confidence. To ensure the high confidence of the PPIs used to generate our network, we set the interaction score threshold to 0.99.

4.4. Prognostic Efficiency and Functions of UDEGs

A survival analysis was performed on the TCGA PDAC cohort using the R package “survival” (version: R 4.0.2)to explore the prognostic efficacy of UDEGs. The CancerMine [51] database was searched to explore the reported roles of the UDEGs in multiple cancers. To analyze the functions of the UDEGs, a KEGG pathway analysis was conducted, and REACTOME [52] (reactome.org, accessed on 20 May 2022) was used to systematically display the hierarchy of the pathway enrichment in this study.

4.5. Processing of scRNA-Seq Data

The raw scRNA-seq data were downloaded from the GSA database (Genome Sequence Archive, ngdc.cncb.ac.cn/gsa, accessed on 21 April 2022, CRA001160 [53]). The “Seurat” [54] (v2.3.0) R package was used for quality control and downstream single-cell analysis. For quality control, we set the following criteria: <200 genes/cell, <3 cells/gene and >10% mitochondrial genes/cell. The tumor and control samples were merged and scaled.

4.6. Cluster Identification and Cell Type Assignment

The top 2000 variable genes were used for clustering. T-SNE reduction was applied to display the clustering results. The cluster-specific marker genes were identified by running the “FindAllMarkers” function of the “Seurat” package. We used some well-known markers from the CellMarker database [55] and the literature to identify specific cell types; for example, we used CPA1, PRSS1, CTRB1, CTRB2 and REG1B to identify acinar cells; MS4A1 and CD27 to identify B cells; AIF1 and CD68 to identify macrophages; PCSK1N, CHGB, CHGA, INS and SLC30A8 to identify endocrine cells; KRT19, MMP7, TSPAN8, SOX9 and LCN2 to identify ductal cells; MUC1 and PROM1 to identify cancer cells; LUM, DCN and LRP1 to identify fibroblasts; MNDA and ITGB2 to identify monocytes; RGS5, ACTA2, PDGFRB and ADIRF to identify stellate cells; and CD2, CD3D and CD3E to identify T cells. Using Seurat’s FindAllMarkers function, we identified genes that were highly expressed in acinar cells, duct cells, macrophages, endocrine cells, fibroblasts, endothelial cells, monocytes, stellate cells and T cells in pancreatic cancer and normal tissues. Then, KEGG functional enrichment analysis was performed on each gene list (p <= 0.05). The functional conformance of common cell types between the pancreatic and normal tissue samples was determined using the following formula:

$$\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{s}\mathrm{t}\mathrm{e}\mathrm{n}\mathrm{c}\mathrm{e}=({\displaystyle \frac{\mathrm{c}\mathrm{o}\mathrm{n}}{\mathrm{n}}}+{\displaystyle \frac{\mathrm{c}\mathrm{o}\mathrm{n}}{\mathrm{t}}})/2$$

For a given cell type, n is the number of pathways enriched with the DEGs of this cell in the normal tissues, t is the number of pathways enriched with the DEGs of this cell in the cancer tissues and con is the number of common pathways in both the cancer and normal tissues.

4.7. Drug Resistance

The processed expression profiles and sensitivity data of 32 pancreatic cell lines were downloaded from the GDSC database (Genomics of Drug Sensitivity in Cancer [56], cancerrxgene.org, accessed on 21 June 2022). The sensitivity of a cell line to a drug is represented by the IC50 value. For each drug–gene pair, we calculated the Spearman correlation coefficient between the expression values of the gene and the IC50 values of the drug across different cell lines. Only the drugs that were effective against at least half of the 32 cell lines and had drug–gene correlations with a p-value < 0.05 were retained.

4.8. Drug–Gene Interaction Network

We explored the potential of the UDEGs to be drug targets by constructing a drug–gene interaction (DGI) network using information from DGIdb [57] (dgidb.genome.wustl.edu, accessed on 23 June 2022). We select drug–gene interactions with scores greater than 5 as candidate interactions and obtained the drugs’ FDA approval information, mechanism and function from the DrugBank database (go.drugbank.com/drugs, accessed on 23 June 2022).

4.9. TMB and MSI Information

The “TCGAmutations” [58] R package was used to calculate the tumor mutation burden (TMB) of the TCGA PDAC, SKCM (skin cutaneous melanoma) and LUSC (lung squamous cell carcinoma) patients. The “TCGAbiolinks” R package was used to download the microsatellite instability (MSI) information of the TCGA patients.

4.10. Immune Cell Proportion of TCGA PDAC Patients

CIBERSORT [59] was applied to obtain the proportions of 22 immune cell types, including macrophages (M0, M1 and M2), CD⁸⁺ T cells, naive CD⁴⁺ T cells, CD⁴⁺ memory resting T cells, CD⁴⁺ memory activated T cells, monocytes, resting dendritic cells, neutrophils, resting mast cells, naive B cells, eosinophils, memory B cells, resting NK cells, activated NK cells, follicular helper T cells, regulatory T cells (Tregs), gamma delta T cells, activated dendritic cells, plasma cells and activated mast cells. The correlations between the expression of the UDEGs and cell proportions were calculated using Spearman correlation.

4.11. Statistical Analysis

All statistical analyses were performed using R 4.0.2.